Adam Bear's last post encouraged me in presenting my own pet hypothesis about the Knobe Effect. It can be found in this paper, written with Emmanuel Dupoux and Pierre Jacob , and forthcoming in Mind & Language. Following Nichols and Ulatowski's proposal, we consider that the Knobe Effect and the Skill Effect can be accounted for only if we accept that "intentionally" has three different meanings that are differently elicited by contextual cues.
To put it in a nutshell, here are (roughly) the three meanings:
1) According to Meaning 1, X is done intentionally only if the agent had a pro-attotude towards X. This meaning is preferentially triggered when X is something we expected the agent to desire (where "expectations" can be normative or statistical, and where normative expectations usually trump statistical expectations)
2) According to Meaning 2, X is done intentionally only if the agent was not reluctant or forced to do X. This meaning is preferentially triggered when X is something we expected the agent to be reluctant to do.
3) According to Meaning 3, X is done intentionally if the agent had control over his action and X was not brought by pure chance or accident. This meaning is preferentially triggered when it is salient that the agent needs skill or luck to perform his action.
Now, what does it have to do with Adam's results? I think that this hypothesis can easily explain those. One thing we insist upon on our paper is that Meanings 1 and 2 are insensitive to how much control the agent has upon his action, while Meaning 3 is very sensitive to this factor. Now, Adam's virtuous scenarios will preferentially elicit Meaning 1 while his reprehensible scenarios will preferentially elicit Meaning 2, leading to Meaning 3 being less employed, and answers be less sensitive to the agent's control upon his action. On the contrary, neutral scenarios will tend to elicit less Meanings 1 and 2 (for lack of expectations), and more Meaning 3, and thus more answers sensible to the agent's control.
This account leads to two further predictions:
1) Intentionality ratings will be much more correlated with the amount of desire attributed to agents in the virtuous and reprehensible scenarios than in the neutral scenarios.
2) The same pattern of results could be found for non-moral scenarios if we contrast three scenarios: (i) scenarios in which participants expect the agent to deeply desire the outcome, (ii) scenarios in which participants expect participants to be deeply reluctant to bring about the outcome, and (iii) scenarios in which the outcome is so indifferent we do not expect the agent to desire or to be reluctant to bring it about. In cases of causal deviances, I predict that intentionaliy ratings will be higher for scenarios (i) and (ii) than for scenarios (iii).